Everything here is optional. The practical rule from the workshop: start with one hosted model, one development workspace, and one revision-control workflow. Add local models or multi-provider routing only when you have a reason.
Models and model ecosystems
- ChatGPT / OpenAI
- General-purpose hosted AI for research, writing, analysis, and coding. A useful reference point for modern model and agent workflows.
- Claude / Anthropic
- Hosted AI widely used for long documents, analysis, and software development. Useful when a project involves substantial context.
- OpenRouter
- Access many model providers through one interface and API. Useful for comparing models, switching providers, and reducing dependence on a single vendor.
- Cohere
- Canadian AI company founded in Toronto, focused on enterprise search, retrieval, reranking, and agent workflows. Especially relevant to knowledge-heavy organizations.
- DeepSeek
- Reasoning and coding models, open-weight releases, and API access. Useful for comparing capabilities, costs, and model ecosystems.
- MiniMax
- Hosted and open-weight multimodal, coding, and agentic models, including long-context options for substantial documents and projects.
Agents and development
- OpenCode
- Open-source coding agent for the terminal, desktop, and IDE. It can connect to many model providers and work directly inside a project.
- Hermes Agent
- Open-source agent from Nous Research with tools, persistent memory, and skills for multi-step work across terminal and desktop workflows.
- Visual Studio Code
- A practical development workspace combining files, an integrated terminal, Git, and extensions, including coding-agent integrations.
- LM Studio
- Desktop software for downloading, running, and serving local models. Useful for experimentation, privacy, and offline workflows.
- Git + GitHub
- Version control and collaboration: track changes, compare revisions, work in branches, and recover earlier versions of a project.
What about open source?
The workshop's open-source slide is a bigger map, and it is on the slides page. It groups proven tools by the kind of work they do: geospatial and mapping (QGIS, OpenStreetMap, OpenDroneMap), data and analytics (Jupyter, Python, PostgreSQL, Superset), farm management and operations (farmOS, ThingsBoard, Node-RED), AI and machine learning (Ollama, LangChain, vLLM), hardware and IoT (Home Assistant, ESPHome, Arduino), and collaboration (Nextcloud, Mattermost, BookStack, Git).
The reason it matters is not ideological. Open tools let you inspect, adapt, and combine existing work, keep your data and infrastructure under your own control, and avoid being locked into a vendor's decisions. You rarely need to start from scratch. The workshop put it well: open source means standing on the shoulders of a global community to go further, together.
How to choose
- Start hosted. You do not need to run anything locally to begin.
- Pick one place to work. A folder plus one agent or assistant is a complete setup.
- Add a second model only to compare, not to feel properly equipped.
- Go local when privacy requires it, or when you want to work offline. Not before.
- Add routing and orchestration last. Two providers and a workflow engine is a project in itself.
Prices, model names, and capabilities move quickly, and any list like this ages within months. Treat it as a starting point for a search, not a current recommendation. The Future Herd Discord is where the group compares what is working right now.